The AI Video Revenue Boom Hides a Brutal Cost Reality
AI video generation is no longer a novelty—it’s a business. Higgsfield, Kling, fal, HeyGen, and Runway collectively disclose revenues totaling roughly $2.3 billion annually, with the bulk of that value captured at the application layer rather than by foundational model vendors. Yet the same week that revenue figures circulated, OpenAI’s Sora app shut down on April 26, reportedly burning around $1 million per day in operating costs. The juxtaposition is stark: the tools to make AI video are thriving, but the business of simply watching AI-generated content remains unproven.
Why the Application Layer Wins
The revenue split tells a clear story. Companies like HeyGen and Runway package generation capabilities into workflows that enterprises will pay for—marketing videos, personalized avatars, product demos. They charge per seat or per render, capturing recurring revenue without bearing the full cost of frontier model training. Model vendors, by contrast, face a brutal unit economics problem: inference is expensive, and consumer demand for raw generation is price-sensitive and sporadic.
Sora’s shutdown underscores the gap between technological capability and sustainable business model. A consumer-facing app that offers free or low-cost generation at scale cannot absorb $1 million daily in compute and serving costs without either massive advertising revenue or a subscription base willing to pay far more than typical consumer AI pricing.
The Cost Paradox Explained
The paradox is that while the marginal cost of generating a single video clip has fallen dramatically—thanks to better models, optimized inference, and competition—the cost of producing a finished, broadcast-quality piece remains stubbornly high. Post-production, editing, sound design, and creative direction still require human expertise. AI accelerates asset creation but does not eliminate the workflow around it.
- Generation cost: Down sharply year-over-year as model efficiency improves.
- Production cost: Remains high due to human oversight, integration, and quality control.
- Distribution cost: Streaming and hosting AI video at scale is non-trivial.
- Monetization gap: Consumers expect free or cheap; enterprises pay for outcomes.
Forward-Looking: The Enterprise Pivot
The path forward for AI video is not consumer entertainment but enterprise tooling. The companies posting real revenue are those selling to businesses that need video for marketing, training, and communication—not those hoping to become the next YouTube of AI-generated content. As Sora’s retreat shows, the consumer attention economy is a harsh master. The next phase of growth will come from embedding AI video into existing paid workflows, where cost is justified by measurable ROI rather than ad-supported eyeballs.




